[ad_1]
Hey @insellup! 👋
Turbo-instruct is a legacy completion model, and it is intended to behave in this manner. If it appeared to be “working” in previous versions, it was actually a mistake on our part— the model wasn’t utilized properly. You can try rolling back to version 2.0.8 and use the turbo-instruct model in the playground; you’ll observe the same behavior occurring (or directly on the OpenAI playground).
So, what you’re encountering in the latest version is actually turbo-instruct. 😺